Recently we have been concerned with locating and tracking targets in aerial videos. Targets in aerial videos usually have weak boundaries due to moving cameras. For the purpose of target detecting, detecting the contour of the target is needed and can help with improving the accuracy of target tracking. Edge detection has assisted in obtaining some advances in this effort. However, noisy images and weak boundary limit the performance of existing contour detecting algorithms. After analyzing the structures and edge maps of a Holistically-nested Edge Detection network, we utilize the highest level side-output and improve the architecture of HED; firstly we cut and resized our images into 400∗320 pixels. Secondly, we detected edges using our improved HED network. Finally, the contour of an object is found based on edge detecting in the previous stage. We have significantly decreased time spent by reducing 5 side output layers to only 1 and replacing the fusion layer with a refinement and image processing module which also helps with the result. The experimental results show that our algorithm outperforms the state-of-the-art regarding images with noise and weak boundary.